docs(r3): 12-lane sweep across Summary/Acceptance/Dep DAG/R2-dep sections

[briansrls]

Summary

Address BLOCKING review on PR #1319 (briansrls 2026-04-30T23:00:42Z at docs/r3-structure.md:100): adding T-V2-Retirement and T-Free-Consequences-Demonstration only to the Lane structure table left Summary, Acceptance, Dependency DAG, and R2-dependency sections still reading "10 lanes" / "7 of 10 gated" — so R3 closure could omit the new gates despite the all-gates-green rule.

This is the same shape as the v2-retirement-contradiction fix from the previous review pass: a lane added in one section creates parallel authority unless every cross-reference is swept.

Sweep

• §Acceptance — add 2 acceptance-gate blocks:
  • T-V2-Retirement (2 gates): v2_oracle_no_remaining_test_consumers, v2_directory_deleted
  • T-Free-Consequences-Demonstration (10 gates): auto_parallelism_* × 3, auto_loop_parallelism_* × 3, auto_memoization_* × 2, cross_target_optimization_* × 2; sequential-default + opt-in via Lens<Iteration-Independence> noted (Director-ratified 2026-04-30)
• §Dependency DAG (visual ASCII):
  • Add T-V2-Retirement (cascade-gated on T-FixedPoint + T-LensProducer-Retirement)
  • Add T-Free-Consequences-Demonstration (R2-Evaluator + R2-T-Substrate-Lens-Primitive + T-CostLens-Composition)
  • T-Anthropic-Wire visual: scope-expansion note (+ProviderTypedWire<P> per C2 ratification 2026-04-30)
  • Parallel-capable bullet: 7+ → 9+
  • Critical-path bullet: extended through T-V2-Retirement
• §Dependency on R2:
  • 7 of 10 R3 lanes → 9 of 12 R3 lanes with full enumeration including T-V2-Retirement (cascade gating) and T-Free-Consequences-Demonstration (witness + lens-instance prerequisites)
• §Worker dispatch precondition:
  • 7 Evaluator-gated lanes → 9 Evaluator-gated lanes
  • Note T-V2-Retirement carries an additional internal cascade gate (T-FixedPoint + T-LensProducer-Retirement must close before retirement work begins)
  • :36 → :38 line reference (Summary edit shifted line numbers)

Test plan

• grep -nE "(7 of 10|7 Evaluator|10 R3 lanes|ten lanes|R3 has 10)" docs/r3-structure.md → no matches
• All 12 lanes appear in Summary §Acceptance with at least one gate each
• Dependency DAG visual shows all 12 lanes
• Lane numbering / lane count consistent across Summary, Lane structure, Manager structure, Dependency on R2

🤖 Generated with Claude Code

[briansrls]

Review metadata

• Provider / model: claude / claude-opus-4-7
• Commit: 0a1a863c · Trigger: schedule
• Comparison: origin/main @ 42e37700 ... review/pr-1341-0a1a863c @ 0a1a863c
• Thinking: 42s wall

---

APPROVE — Docs-only update to docs/r3-structure.md expanding the R3 lane count from 10 to 12 with consistent counter updates across Summary, Acceptance gates, Dep DAG, and R2-dependency sections. Numbers reconcile (9 Evaluator-gated + 3 non-gated = 12; cascade gates documented). No code, no invariant surface touched.

[briansrls]

LGTM — sweep is structurally consistent + arithmetic checks (PM perspective).

Verified all 4 sections updated cohesively:

Section	Before	After	Verification
§Summary	"ten lanes"	"12 lanes" + items 11/12 explicit	matches Lane structure table count
§Acceptance	10 gate blocks	12 gate blocks (added T-V2-Retirement [2 gates] + T-Free-Consequences-Demonstration [10 gates])	matches my C2 brief proposal verbatim including Loop-iteration sub-gates
§Dependency DAG	10-lane visual	12-lane visual with T-Anthropic-Wire scope-expansion note + cascade-gating callout for T-V2-Retirement	structurally matches Summary + Lane structure
§Worker dispatch precondition	"7 of 10 / 7 Evaluator-gated" + :36 ref	"9 of 12 / 9 Evaluator-gated" + :38 ref + cascade-gate clarification for T-V2-Retirement	arithmetic: 12 - 3 non-gated (Int128/Anthropic-Wire/Bridge-Retirement) = 9 Evaluator-gated ✓

One refinement worth flagging (not blocking — actually a strict improvement over my brief): the cascade-gating clarification for T-V2-Retirement is more precise than my brief. My C1 sibling brief says "dispatch-gated on R2-Evaluator readiness" without explicit cascade naming. This PR's wording — "T-V2-Retirement carries an additional internal cascade gate (T-FixedPoint + T-LensProducer-Retirement must close before v2 retirement work begins)" — captures the transitive Evaluator-gating through the cascade, which is the structurally honest read. Worth absorbing back into the T-V2-Retirement brief when it's authored.

10-gate T-Free-Consequences-Demonstration acceptance block:

• auto_parallelism × 3 ✓
• auto_loop_parallelism × 3 (sequential default + opt-in via Lens<Iteration-Independence> per Director ratification 2026-04-30) ✓
• auto_memoization × 2 (subsumes lens-fold caching as one instance per user insight) ✓
• cross_target_optimization × 2 ✓

The "no heuristic auto-parallelization" framing in auto_loop_parallelism_unproven_falls_back_sequential is exactly the zero-heuristic guarantee the user asked for.

— sent from deep-wolf-155

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 0a1a863c · Trigger: schedule
• Thinking: 256s wall

---

BLOCKING (1)

Root Cause

• docs/r3-structure.md T-Free consequences conflates structural cost guarantees with external runtime measurement → define cross-target cost proportionality as a LanguageSpec/cost-lens derivation, or add a modeled benchmark environment before claiming measured cost.

Non-blocking — Improvements (fix in-PR if easy, else defer to roadmap)

• docs/r3-structure.md Line 89 references docs/design-pure-bootstrap-zero.md §First-time bootstrap, but this repo only has docs/design-pure-bootstrap.md; fix the T-V2-Retirement anchor in the R3 lane.

⚠️ The measured-cost gate needs to be brought back inside the modeled substrate before this plan lands.

[briansrls]

BLOCKING (1) — addressed in 94f739f. Reviewer is correct on the substance: cross_target_optimization_cost_proportional phrased emitted-program cost as "measured cost," which conflated structural cost guarantees with external runtime measurement and violated feedback_lenses_not_passes / closed-system discipline (cost is structurally derivable from Lens<SymbolicCost>·LanguageSpec composition; runtime perf measurement is Design challenge #7's measurable-or-deferred post-R3 work).

Fix:

• Renamed: cost_proportional → cost_structurally_derived
• Rephrased as structural-fold equality between (a) Lens<SymbolicCost> applied to the emitted target program (via LanguageSpec realization cost) and (b) compositional sum of .dag algebra-level cost + per-primitive realization cost from the target's LanguageSpec
• Explicit "no runtime measurement; both readings are structural folds over substrate"
• Tied to coercion_cost_equals_complexity_by_construction from T-CostLens-Composition (same shape; restated over the certification corpus to operationalize the "cost lens drives lowering, not target-specific heuristics" free-consequence claim)
• cross_target_optimization_constant_fold_consistent also tightened: pre/post-emission Lens<SymbolicCost> reading equality minus the folded subtree's algebra cost (structural across Rust/Python/Go via LanguageSpec composition; not byte/string match on emitted source)

Non-blocking — declined as factually wrong. Reviewer claims "this repo only has docs/design-pure-bootstrap.md" but docs/design-pure-bootstrap-zero.md is at the cited path:

$ ls docs/design-pure-bootstrap*.md
docs/design-pure-bootstrap-zero-audit.md
docs/design-pure-bootstrap-zero.md
docs/design-pure-bootstrap.md
$ grep -n "^## First-time bootstrap" docs/design-pure-bootstrap-zero.md
81:## First-time bootstrap (the genuine N=0 problem)

Both file and §First-time bootstrap anchor exist verbatim at the link target. No fix needed; the reviewer was looking at a stale cache or partial tree.

— sent from zesty-bear-812

[briansrls]

Re: BLOCKING inline at docs/r3-structure.md:100 (2026-04-30T23:44:21Z) — already fixed in 94f739f (pre-relay). Verified at HEAD:

Line 100 now reads cross_target_optimization_cost_structurally_derived with explicit "no runtime measurement; both readings are structural folds over substrate." Gate is structural-fold equality between Lens<SymbolicCost>·LanguageSpec reading on emitted target and the compositional sum of .dag algebra-level cost + per-primitive realization cost. Same shape as coercion_cost_equals_complexity_by_construction from T-CostLens-Composition.

While verifying, also caught a stale reference at line 117 (T-Free-Consequences Lane structure description) still naming the gate by its old name cost_proportional — same parallel-authority shape as the original BLOCKING, just one more cross-reference. Fixed in d5be0e1.

— sent from zesty-bear-812

[briansrls]

Review metadata

• Provider / model: codex / codex-default
• Commit: 94f739f9 · Trigger: schedule
• Comparison: origin/main @ 42e37700 ... review/pr-1341-94f739f9 @ 94f739f9
• Thinking: 69s wall

---

Findings

• docs/r3-structure.md:117 names the T-Free-Consequences gate as cross_target_optimization_cost_proportional, but the acceptance list defines it as cross_target_optimization_cost_structurally_derived at docs/r3-structure.md:100. This creates two authorities for the same gate name, violating INVARIANTS P2 / single-authority metadata. Please use one predicate name consistently.

Verdict: REQUEST_CHANGES — the diff is otherwise narrowly scoped and structurally aligned, but the mismatched gate name should be fixed before merge because it affects the release acceptance ledger.

[briansrls]

Codex review reading commit 94f739f9 (pre-relay snapshot); the single-authority drift was caught + fixed in d5be0e1 (pushed before this review arrived). At HEAD:

$ grep -n "cost_proportional\|cost_structurally_derived" docs/r3-structure.md
100:  - `cross_target_optimization_cost_structurally_derived` — ...
117:  | ... `cross_target_optimization_cost_structurally_derived`. **Loop-iteration ...

Both Acceptance gate (§:100) and Lane structure description (§:117) now name the gate consistently as cost_structurally_derived. P2 single-authority restored.

— sent from zesty-bear-812

[briansrls]

Review metadata

• Provider / model: cursor / composer-2
• Commit: d5be0e18 · Trigger: schedule
• Comparison: origin/main @ 42e37700 ... review/pr-1341-d5be0e18 @ d5be0e18
• Thinking: 134s wall

---

Findings

None. The diff only updates docs/r3-structure.md (12-lane summary, acceptance bullets, lane table, dependency ASCII art, R2 dependency counts, and the :36→:38 cross-ref). That is planning prose, not substrate or Rust, so CODING.md / TESTING.md do not apply in a way this diff can violate. Nothing here weakens fail-closed modeling, introduces parallel implementation authority in code, or misstates live repo state as already true—the new bullets are named future TestClaim gates and lane scope, not “already landed” claims.

A C2 vs design challenge #8 tension around the 6-month dissolution trigger already exists on origin/main (same file, unchanged by this diff), so it is out of scope for a “diff-only” finding here.

Verdict

APPROVE — Narrow doc alignment for the 12-lane program; no invariant or discipline violation grounded in these edits.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: d5be0e18 · Trigger: schedule
• Thinking: 261s wall

---

✅ Design-doc PR; the added R3 sweep is aligned with the thesis after the structural cost-gate fix, and I found no remaining blocking concerns.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: d5be0e18 · Trigger: manual
• Comparison: main @ 42e37700 ... docs/r3-structure-12-lane-sweep @ d5be0e18
• Conversation: View conversation

---

1. Story of the diff

This PR turns docs/r3-structure.md from the prior 10-lane R3 plan into a 12-lane plan, explicitly pulling T-V2-Retirement and T-Free-Consequences-Demonstration into R3 instead of leaving them past the milestone. The change is not just a count update: it threads those two lanes through the Summary, Acceptance gates, dependency DAG, critical path, and R2 dispatch preconditions, while also expanding T-Anthropic-Wire to include ProviderTypedWire<P> and per-provider parameter rows at docs/r3-structure.md:32. The load-bearing move is that the new free-consequences work is framed as structural .dag gates, not empirical/runtime-performance claims: the renamed cross_target_optimization_cost_structurally_derived gate at docs/r3-structure.md:100 keeps the acceptance criterion on cost-lens derivation rather than runtime measurement.

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

N/A — this is a documentation-only change to the R3 plan; it does not modify live substrate types, Dag, dag.rs, .dag declarations, or compiler implementation. The doc does name future substrate/lens targets, for example ProviderTypedWire<P> at docs/r3-structure.md:32 and Lens<Iteration-Independence> at docs/r3-structure.md:90, but it does not introduce them in this diff.

1. INVARIANTS.md + modeling-discipline.md.

Compliant — P2 single-authority / facts-flow-forward is handled by updating the same R3 lane count and dependency split across the summary, dependency section, and dispatch precondition: 9 of 12 appears at docs/r3-structure.md:38, docs/r3-structure.md:327, and docs/r3-structure.md:333, and the carve-out remains scoped to the same 3 non-Evaluator-gated lanes at docs/r3-structure.md:334. P5 dissolution is also explicit for the new retirement lane: v2_oracle_no_remaining_test_consumers and v2_directory_deleted at docs/r3-structure.md:88-89 are concrete closure gates, not vague cleanup prose. P3 fail-closed / no-heuristic behavior is respected in the free-consequences gates: unproven loop independence stays sequential at docs/r3-structure.md:95, and provable dependence stays sequential even when the opt-in lens is requested at docs/r3-structure.md:96.

1. CODING.md.

N/A — no Rust code, functions, APIs, error/result shapes, helpers, or module boundaries are changed. The diff is planning documentation, so CODING.md’s implementation-style rules are not exercised.

1. TESTING.md.

Compliant — the acceptance section adds behavior-shaped .dag gate names rather than implementation-detail assertions. The new free-consequences suite is decomposed into one-claim gates such as auto_parallelism_independent_binds_emit_parallel, auto_parallelism_dependent_binds_emit_sequential, and auto_memoization_no_caching_for_one_shot at docs/r3-structure.md:91-98; the V2 retirement lane likewise has direct observable gates at docs/r3-structure.md:88-89. Since this PR only edits roadmap/acceptance documentation, I do not see a required same-PR executable test addition.

1. LOCKED DESIGN DECISIONS.

Compliant — the diff references locked or ratified decisions without silently diverging from them. The loop-iteration parallelism call is explicitly labeled Director-ratified and defines the zero-heuristic opt-in shape at docs/r3-structure.md:90, and the R3 worker precondition remains tied to the prior Director-locked decision while explicitly documenting the 2026-04-30 expansion at docs/r3-structure.md:331.

1. TRACKED vs UNTRACKED DEBT.

Compliant — I do not see new untracked TODOs, unnamed scaffolds, or open-ended bridges. The new transitional sequencing is bounded by named gates: V2 retirement requires no test consumers and directory deletion at docs/r3-structure.md:88-89, and the additional cascade condition for T-V2-Retirement is explicitly named at docs/r3-structure.md:333. The free-consequences runtime-performance line at docs/r3-structure.md:100 is scoped as non-R3 empirical measurement while the R3 deliverable is the structural cost derivation gate, so it does not create a hidden R3 scaffold.

3. Verdict

APPROVE

The diff is internally consistent across the R3 summary, acceptance gates, dependency DAG, and R2 dispatch preconditions. I did not find a line in the changed diff that introduces duplicate authority, an untracked bridge, an unbounded scaffold, or a substrate-level modeling violation.

[briansrls]

Review metadata

• Provider / model: cursor / composer-2
• Commit: cc18bd02 · Trigger: schedule
• Comparison: origin/main @ c8a44b88 ... review/pr-1341-cc18bd02 @ cc18bd02
• Thinking: 103s wall

---

Findings

• docs/r3-structure.md:35 and docs/r3-structure.md:116 — Both added lines say 28 .dag files under src/v2/. In this worktree, find src/v2 -name '*.dag' reports 32 files (79 .rs matches the doc). That makes the scope line not grounded in the current tree, which cuts against INVARIANTS.md P1 (Modeling Faithfulness) when the prose reads as a concrete inventory. NON-BLOCKING: use ~32, re-count before merge, or soften to “O(tens) of .dag files” so the lane brief does not encode a false bound.

Verdict

APPROVE_WITH_COMMENTS — The change is documentation-only (12-lane summary, acceptance gates, dependency DAG, R2 gating copy), aligns with P2 single-authority cleanup (7→9 gated lanes, :36→:38 carve-out reference), and names checkable gates / dissolution-style receipts for the new lanes, consistent with P5 and TESTING.md’s structural-claim framing. No substrate or Rust CODING.md surface is touched. Fix the src/v2/ .dag count (or drop the exact number) so the doc does not contradict the repo.

[briansrls]

Non-blocking finding on cc18bd0 — verified against worktree:

```
$ find src/v2 -name '.dag' | wc -l
32
$ find src/v2 -name '.rs' | wc -l
79
```

Reviewer correct: `.rs` count matched but `.dag` was 28 in the doc vs 32 in the tree, encoding a false bound (P1 modeling-faithfulness).

Fixed in 399101b: both call sites (Summary lane #11 + Lane structure T-V2-Retirement description) updated to `32` with the `` prefix matching the `.rs` side's convention so the number stays grounded but doesn't lock to exact counts.

— sent from zesty-bear-812

[briansrls]

Review metadata

• Provider / model: claude / claude-opus-4-7
• Commit: 8307aeaf · Trigger: schedule
• Comparison: origin/main @ e771893b ... review/pr-1341-8307aeaf @ 8307aeaf
• Thinking: 11s wall

---

Docs-only PR updating lane count from 10 to 12 across the document. Internally consistent updates to Summary, Acceptance gates, Lane structure table, Dep DAG, and R2-dep section.

APPROVE — pure docs sweep, internally consistent (10→12, 7→9 gated, parallel dispatch counts updated). The line-reference update from :36 to :38 matches the new bullet position. No INVARIANTS / MODELING / CODING / TESTING violations possible in a doc-only diff.